It’s become a fashionable sub-genre of sports content publishing to create articles based on AI predictions. In the era before ChatGPT and Gemini, you’d often see similar articles titled “Supercomputer predicts X,” but now it is common to see content using proprietary AI models and chatbots, like ChatGPT, to churn out the content.
Right now, in the middle of the NFL Playoffs, there’s a plethora of this content online, produced by large mainstream media organizations and small blogs alike. While we will have to wait until we can see with hindsight how well different AI models do across the Playoffs and Super Bowl, we can look back at some of the AI predictions from preseason. And it sure does look like there is work to be done before we start believing that AI is some sort of omnipotent cheat code in sports predictions.
AI Models Flopped On Ravens And Chiefs Picks
A case in point was Sports Radar’s AI model, which was used to simulate several Super Bowl outcomes. The model settled on the Baltimore Ravens as the most likely Super Bowl champion. Yes, the same Baltimore Ravens that went 8-9 and never made the Playoffs. Second on Sports Radar’s list was the Philadelphia Eagles, who were eliminated over the Wild Card Weekend. The Buffalo Bills were third, and they are still in the Playoffs at the time of writing.
Now, none of this is meant to criticize Sports Radar, an excellent sports data company. Rather, it’s to point out that AI, in general, is not yet up to scratch yet when making these sports predictions. In fact, the AI models feel just as fallible as humans, perhaps more so. Plenty of other publications pulled out AI predictions citing the Chiefs (eliminated early in the regular season) and Lions (failed to make Playoffs) as big Super Bowl contenders.
Super Bowl Betting Odds Have Changed
The problem, perhaps, is that AI is looking too much at the previous season for clues to the next one. Most of the AI models’ picks were in the Playoffs last season. But we know that NFL Playoffs and Super Bowl odds chop and change over the course of the season, and there are always surprise packages. Teams like the Seattle Seahawks and New England Patriots came to the fore this season. We have yet to see any AI models that predicted those two teams would have 14-3 seasons and share the best records in the NFL.
What is clear, though, is that AI models will closely resemble the betting odds preseason, and the betting odds will, in turn, reflect the consensus picks from sports pundits and media outlets. Of course, now and then, you’ll find a savvy human expert who predicted that a certain team would defy the odds and consensus, but AI models will largely follow the consensus.
Limit On What AI Can Do With Sports Data
In the simplest terms, AI is largely used to find that consensus, usually from multiple sources – say, ESPN, CBS Sports, USA Today, and so on. It will, of course, have access to other data points, and it may be able to make calculations based on the possible impact of big trades and arrivals from the draft, but there is a limit. In this case, human analysts are probably still better at anticipating how players will fit, even if AI models can do much more analysis of the underlying numbers.
In the end, this is simply sports. We started this article talking about how AI performed poorly in making predictions preseason, but just about every analyst and sportsbook did the same. It is an undisputable fact that the preseason favorite for the Super Bowl rarely goes on to win the big game. There are always surprise packages and disappointments. AI might improve over time, in sports as well as other fields, but there will always be a barrier to perfection in sports predictions.
